An algorithm for network and data-aware placement of multi-tier applications in cloud data centers

Abstract

Today's Cloud applications are dominated by composite applications comprising multiple computing and data components with strong communication correlations among them. Although Cloud providers are deploying large number of computing and storage devices to address the ever increasing demand for computing and storage resources, network resource demands are emerging as one of the key areas of performance bottleneck. This paper addresses networkaware placement of virtual components (computing and data) of multi-tier applications in data centers and formally defines the placement as an optimization problem. The simultaneous placement of Virtual Machines and data blocks aims at reducing the network overhead of the data center network infrastructure. A greedy heuristic is proposed for the on-demand application components placement that localizes network traffic in the data center interconnect. Such optimization helps reducing communication overhead in upper layer network switches that will eventually reduce the overall traffic volume across the data center. This, in turn, will help reducing packet transmission delay, increasing network performance, and minimizing the energy consumption of network components. Experimental results demonstrate performance superiority of the proposed algorithm over other approaches where it outperforms the stateof-the-art network-aware application placement algorithm across all performance metrics by reducing the average network cost up to 67% and network usage at core switches up to 84%, as well as increasing the average number of application deployments up to 18%.

Optimizing multi‐tier application performance with interference and affinity‐aware placement algorithms

“…An overloaded network impacts the cloud in several ways, such as increase of infrastructure costs and reduction of Quality of Service (QoS). 4 Moreover, as multi-tier applications are composed of several individual applications and as clouds host multiple applications in the same PM, resource interference may be another factor of concern. Therefore, given these problems, it is not only important for cloud providers to use techniques to reduce network traffic, placing applications close to the data they need, but it is also important to avoid placing applications that use the same resource intensively in the same PM.…”

“…With the rise of applications running on the cloud, including the multi‐tier ones, cloud data centers have become more concerned with network utilization. An overloaded network impacts the cloud in several ways, such as increase of infrastructure costs and reduction of Quality of Service (QoS) . Moreover, as multi‐tier applications are composed of several individual applications and as clouds host multiple applications in the same PM, resource interference may be another factor of concern.…”

Rack Aware Data Placement for Network Consumption in Erasure-Coded Clustered Storage Systems

“…The data is sent to the sink node through different parent nodes at a certain probability, which can reduce transmission congestion and the end-to-end delay. At the same time, in Reference [32], a greedy heuristic algorithm is proposed for the on-demand application components placement, which localizes network traffic in the data center interconnect. This optimization helps reduce communication overhead in upper layer network switches and will eventually reduce the overall traffic volume across the data center.…”

Decision Making Framework for Decentralized Virtual Machine Placement in Cloud Computing

“…The authors in [9], discussed VM migration using the maximum processing power (MPP) and random selection (RS) approach AH, the solution proposed by [9] preserved same firewall rule for VM to the destination server after migration. The authors in [10], have proposed Hierarchical Decentralized Dynamic VS Consolidation Framework, where they discussed how the global controller does take decision for VM migration.…”

“…The author [10] has considered the future utilization of the server. The servers in the solution proposed by the author [10] do not consider decisions from another server before taking decision for VM migration.…”